**1. Introduction**

Necrotizing enterocolitis (NEC), defined by the Bell classification [1–3], is the most severe intestinal disease in preterm infants, with a mortality score of 25% and long-term neurological morbidity [4]. Yet, a precise initiating factor of this pathology is missing. In the last decade, gu<sup>t</sup> microbiota was identified and recognized as a specific organ with functions widely beyond digestion [5]. Both its taxonomic (relative abundance) and functional (microbial pathway) alterations, named dysbiosis, were described in several pathologies, in particular, metabolic diseases such as type 2 diabetes and obesity [6–8], and intestinal inflammatory diseases [9,10]. Importantly, a dysbiotic gu<sup>t</sup> microbiota associated with a very high inflammatory status of the gu<sup>t</sup> [11,12] may trigger NEC development, since germ-free mice do not develop NEC [13].

From a clinical and microbiological point of view, studies of NEC were focused only on established and severe phenotypes, such as NEC-2 and NEC-3. Based on the French study Etude épidémiologique sur les petits âges gestationnels (EPIPAGE 2), the incidence of proved NEC-2 and NEC-3 is 1–5% in preterm infants born at less than 32 weeks of gestation [14].

By contrast, NEC suspicions such as lethargy, bradycardia, thermic instability associated with biliary gastric residues, vomiting, abdominal distension with or without rectal bleeding, with a normal abdominal x-ray image or a simple dilatation, which identifies suspected NEC (NEC-1), have not been studied yet. In fact, enteropathies are frequent in the first weeks of life in preterm infants, though no data are available about NEC-1 incidence. This induces the end of alimentation, a prolonged (sometime life-lasting) parenteral nutrition, with a delayed gu<sup>t</sup> maturation and failure to thrive [15]. Therefore, to study the evolution of gu<sup>t</sup> microbiota and microbiome during the early onset of NEC, we focused on NEC-1 children within the first two months of life. We studied the fecal metabolome, to understand how a change in gu<sup>t</sup> microbiota may drive alterations in intestinal metabolites. To further understand which factor of mother and child may a ffect the evolution of gu<sup>t</sup> microbiota, microbiome and fecal metabolome during NEC-1, we analyzed: the presence of neonatal antibiotherapy (ABx), ABx treatment on the mother, childbirth (Cesarean-section (C-sec) vs. vaginal birth (VB)), very low birth weight (VLBW), extreme low birth weight (ELBW) and gestational age (GA) > or ≤ 28 weeks.

#### **2. Research Design and Methods**

### *2.1. Cohort Constitution*

We conducted a prospective monocentric case-control cohort study. This study was approved (number of the approval: DC 2016-2804) by Neonatal and Pediatric Intensive Care Unit and Neonatology Department of Purpan Hospital in Toulouse, France. The parents of the children involved in this study gave their approval by written consent. The inclusion criteria regarding all of the children hospitalized into the Neonatal and Pediatric Intensive Care Unit or Neonatology Departments of the Purpan Hospital, were:


Following the inclusion of every case, we conducted in parallel a search for two controls, according to the following matching criteria, listed in decreasing priority:


Inclusion criteria for controls were:


Children with complex congenital cardiopathy or with spontaneous intestinal perforation without a radiological evidence of NEC were excluded from the study.

Based on these criteria, we included 11 NEC-1 children, with 27 feces collection (4 fecal samples for time-point 1–10 days (d); 10 fecal samples for time-point 11–20 d; 7 fecal samples for time-point 21–30 d; 6 fecal samples for time-point > 30 d) and 21 healthy children, with 53 feces collection (15 fecal samples for time-point 1–10 d; 14 fecal samples for time-point 11–20 d; 13 fecal samples for time-point 21–30 d; 11 fecal samples for time-point > 30 d). Hence, a total of 80 fecal samples was analyzed in our study. The period of collection was day 1 to day 68 of life of the newborn.

#### *2.2. Taxonomic and Functional Analysis of Gut Microbiota*

Feces analyzed in this study were collected by nurses in the related department in the first week of life and once a week till the end of the hospitalization. Feces were firstly kept at 4 ◦C in a 5 mL Eppendorf tube with 20% glycerol/Lysogeny Broth and then stored at −80 ◦C. Total DNA was extracted from feces as previously described [16], with a modification: a thermic shock of 30 seconds was performed between each bead-shaking step (3 bead-shaking steps of 30 seconds each at maximum speed). The 16S bacterial DNA V3–V4 regions were targeted by 357wf-785R primers and analyzed by MiSeq (RTLGenomics, Texas, USA). An average of 68,669 sequences was generated per sample. Bioinformatic filters applied as already described [17]. Cladogram and LDA scores were drawn via the LEfSe algorithm [18]. Diversity indices were calculated using the software Past 4.02 (Hammer, Ø., Harper, D.A.T., and P. D. Ryan, 2001. PAST: Paleontological Statistics Software Package for Education and Data Analysis. Palaeontologia Electronica 4(1): 9pp). The predictive functional analysis of the gu<sup>t</sup> microbiota was performed via PICRUSt [19]. Diseases and host genetic variation linked to NEC-1\_21–30 d-associated gu<sup>t</sup> microbiota were identified via MicrobiomeAnalyst [20], with the Taxon Set Enrichment Analysis module.

#### *2.3. Fecal Metabolome Analysis*

The metabolome (total metabolites) analysis of the feces was performed as previously described [17]. Fecal samples have been prepared as it follows: 50 mg of feces were homogenized for 30 seconds in 500 μL of a pH 7.0 phosphate bu ffer, prepared in D2O. Then, the homogenate was chilled into ice for 1 minute and centrifuged at 12,000 RPM for 10 minutes at 4 ◦C. The supernatant was then recovered, and the pellet was re-homogenized again at the same conditions of 12,000 RPM for 10 minutes at 4 ◦C. All the supernatants were then pooled and centrifuged at 18,000 RPM for 30 minutes at 4 ◦C. The supernatant was recovered and centrifuged again at the same conditions of 18,000 RPM for 30 minutes at 4 ◦C. A total of 600 μL of the final supernatant was then analyzed into nuclear magnetic resonance (NMR) tubes of 5 mm of diameter. The conformity criterium to validate the final sample was the aspect, to be crystal clear.

Pathway-associated metabolite sets and SNP-associated metabolite sets (Supplementary Figure S3 C,D, Supplementary Figure S4 B,C and Supplementary Figure S6 G) were analyzed via MetaboAnalyst 4.0 [21], with the enrichment analysis module.

### *2.4. Statistical Analysis*

The results are presented as mean ± SEM for histograms and box and whiskers graphs. Statistical analyses were performed by two-way analysis of variance (ANOVA) followed by a two-stage linear step-up procedure of Benjamini, Krieger and Yekutieli to correct for multiple comparisons by controlling the false discovery rate (<0.05) (for histograms) or the Mann-Whitney test (for box and whiskers), as indicated in the figure legend, by using GraphPad Prism version 7.05 for Windows Vista (GraphPad Software, San Diego, CA, USA). For Table 1, results are presented as median or as indicated and *P* value was calculated using Fisher's exact test. Significant values were considered starting at *P* < 0.05. For the taxonomical and predictive functional analysis of gu<sup>t</sup> microbiota, significant values were considered, starting at *P* < 0.05 or *P* < 0.01 when indicated. Principal component analysis (PCA) graphs were drawn by using Past 4.02.


#### **Table 1.** Cohorts characteristics.
